Applying Machine Learning and RPA to Automate Financial Crime Investigations

Machine learning, a powerful subset of artificial intelligence, and robotic process automation (RPA) have shown promise in addressing various automation opportunities that can benefit the financial services industry. In the Bank Secrecy Act/anti-money laundering world, financial institutions (FIs) are burdened with compliance requirements and growing costs. This has led compliance teams to investigate how to automate manual processes in their anti-money laundering (AML) programs with new technologies.

However, there is much confusion between the role and capabilities of RPA and machine learning technologies and how they are applied. A quick refresher on RPA and machine learning is in order.

RPA technology utilizes software robots, or “bots,” to gather data and perform other repetitive tasks. Within the financial services compliance industry, it has been used to streamline the movement of data between systems, thereby reducing the “swivel-chair” problem for many compliance professionals. Specifically, the technology is being successfully deployed for collecting and validating customer screening data for know your customer purposes, automating new customer onboarding and more.

Machine learning enables FIs to analyze massive amounts of data, find patterns in that data and quickly identify where exceptions or anomalies exist. These technologies have been used to automate workloads associated with investigation processes. Notably, offerings that leverage these emergent technologies have been proven to reduce false positives from transaction monitoring systems, streamline suspicious activity report filing and decrease risk by finding financial crimes that current processes miss.

Within AML, RPA can be employed to gather and organize data for investigators. RPA bots replace manual processes of gathering data from various database sources. However, while RPA can automate the job of grabbing and assembling data for investigators, it would still be incumbent upon compliance teams to analyze every letter and number to determine the relevance of that data within a unique investigation. For example, RPA can be utilized to find an entity included in a piece of adverse media and serve that news article up to an investigator, but it is unable to discern if the entity mentioned in the article is a bad actor. Articles discovered and served up by a bot would still need to be read and evaluated by an investigator simply to find the relevance of the entity to the investigation.

Automating complex analysis of multitudes of data is where machine learning shines. The technology can analyze the data that is collected, as well as identify and provide relevance regarding the risk associated with what was found. What is more, by using related tools such as natural language processing (NLP), machine learning can report why transactions or entities are considered suspicious and even weigh or explain the significance of that suspicious activity. Returning to the adverse media example above, machine learning is being used to determine if the entity is merely mentioned in a piece of adverse media or if that entity is a bad actor in the story. By automating this analysis, machine learning takes care of a laborious and time-consuming aspect of the AML investigation and provides the professional investigator with accurate and relevant information needed to make a decision regarding the case.

While RPA bots are highly effective in gathering data for investigators, the efficiency gains they offer are limited because they cannot provide decision-ready reporting back to time-strapped AML teams. FIs should consider utilizing RPA in conjunction with machine learning technology to automate complex processes with higher accuracy and efficiency.

The following are a few ways in which FIs can automate their AML compliance programs:

  1. Build a solution in-house: Like any build versus buy decision, the downside to this approach is the difficulty in securing a top team of data science experts to build the first version of the system followed by the management of an ongoing maintenance program to ensure success.
  2. Hire a consultant: This approach requires time and focus from internal teams and a big budget for both the consultant and their subcontracted data scientist. Like the in-house approach—to be kept current—data scientists must be hired internally or consultants must be regularly re-engaged.
  3. Combine multiple point solutions: Some vendors offer machine learning point solutions or RPA bots dedicated to addressing very specific individual components of AML challenges such as entity resolution and adverse media screening. Implementing two or three independent solutions still leaves large productivity gaps in processes while unforeseen conflict between point solutions can create entirely new and unpredictable issues.
  4. Implement a packaged solution: Packaged solutions purpose-built for AML that marry the data-gathering strengths of RPA with machine learning’s analytical capabilities deliver tremendously effective and efficient AML outcomes. These solutions should be easy to deploy and can be operational within weeks. Born with financial crime expertise built in, they benefit from continuous updates based on the experience and needs of all the FIs using them.

No matter the manner in which it is accomplished, by employing both RPA and machine learning in AML compliance programs, FIs can automate the investigative process to improve efficiency and reduce costs. Automation also increases the consistency of adjudication and reporting―pleasing regulators while making the day-to-day work of investigators more rewarding as they can spend more of their time on challenging work.

David McLaughlin, founder and CEO, QuantaVerse, PA, USA,

Leave a Reply